Pre-training Vision Models for the Classification of Alerts from Wide-field Time-domain Surveys

Rehemtulla, N ORCID: 0000-0002-5683-2389, Miller, AA ORCID: 0000-0001-9515-478X, Walmsley, M ORCID: 0000-0002-6408-4181, Shah, VG ORCID: 0009-0009-1590-2318, Jegou du Laz, T ORCID: 0009-0003-6181-4526, Coughlin, MW ORCID: 0000-0002-8262-2924, Sasli, A ORCID: 0000-0001-7357-0889, Bloom, J ORCID: 0000-0002-7777-216X, Fremling, C ORCID: 0000-0002-4223-103X, Graham, MJ ORCID: 0000-0002-3168-0139, Groom, SL ORCID: 0000-0001-5668-3507, Hale, D ORCID: 0000-0002-4662-122X, Mahabal, AA ORCID: 0000-0003-2242-0244, Perley, DA ORCID: 0000-0001-8472-1996, Purdum, J ORCID: 0000-0003-1227-3738, Rusholme, B ORCID: 0000-0001-7648-4142, Sollerman, J ORCID: 0000-0003-1546-6615 and Kasliwal, MM ORCID: 0000-0002-5619-4938 (2026) Pre-training Vision Models for the Classification of Alerts from Wide-field Time-domain Surveys. Publications of the Astronomical Society of the Pacific, 138 (3). ISSN 0004-6280

Preview

Text Pre-training Vision Models for the Classification of Alerts from Wide-field Time-domain Surveys.pdf - Published Version Available under License Creative Commons Attribution. Download (9MB) | Preview

Abstract

Modern wide-field time-domain surveys facilitate the study of transient, variable and moving phenomena by conducting image differencing and relaying alerts to their communities. Machine learning tools have been used on data from these surveys and their precursors for more than a decade, and convolutional neural networks (CNNs), which make predictions directly from input images, saw particularly broad adoption through the 2010s. Since then, continually rapid advances in computer vision have transformed the standard practices around using such models. It is now commonplace to use standardized architectures pre-trained on large corpora of everyday images (e.g., ImageNet). In contrast, time-domain astronomy studies still typically design custom CNN architectures and train them from scratch. Here, we explore the effects of adopting various pre-training regimens and standardized model architectures on the performance of alert classification. We find that the resulting models match or outperform a custom, specialized CNN like what is typically used for filtering alerts. Moreover, our results show that pre-training on galaxy images from Galaxy Zoo tends to yield better performance than pre-training on ImageNet or training from scratch. We observe that the design of standardized architectures are much better optimized than the custom CNN baseline, requiring significantly less time and memory for inference despite having more trainable parameters. On the eve of the Legacy Survey of Space and Time and other image-differencing surveys, these findings advocate for a paradigm shift in the creation of vision models for alerts, demonstrating that greater performance and efficiency, in time and in data, can be achieved by adopting the latest practices from the computer vision field.

Item Type:	Article
Uncontrolled Keywords:	5107 Particle and High Energy Physics; 5101 Astronomical Sciences; 51 Physical Sciences; Networking and Information Technology R&D (NITRD); Bioengineering; Machine Learning and Artificial Intelligence; 0201 Astronomical and Space Sciences; Astronomy & Astrophysics; 5101 Astronomical sciences; 5107 Particle and high energy physics
Subjects:	Q Science > QB Astronomy Q Science > QC Physics
Divisions:	Astrophysics Research Institute
Publisher:	IOP Publishing
Date of acceptance:	11 March 2026
Date of first compliant Open Access:	23 April 2026
Date Deposited:	23 Apr 2026 14:08
Last Modified:	23 Apr 2026 14:08
DOI or ID number:	10.1088/1538-3873/ae50bc
URI:	https://researchonline.ljmu.ac.uk/id/eprint/28442

View Item